Fusion of Panchromatic and Multispectral Remote Sensing Data Using an Improved Region Regression Model

Abstract

A region-based regression model is presented to fuse a pair of panchromatic (PAN) and multispectral image (MS) in order to get a pansharpened image with high spectral fidelity. Firstly, the images from two different sensors are segmented into isolated regions respectively by the mean shift segmentation method with embedded edge confidence. Then, three segmentation maps got from two images are used as reference of the following segment-based modeling process. The regression model is established independently for each isolated region by considering the linear relationship between corresponding PAN and MS image pixel values in the region. Finally, a modulation-based fusion technique is employed to get the fusion result. Two pairs of images are used for evaluating the method in both qualitative and quantitative ways. The method shows the superior result.